More complexity creep in Phobos

[Andrei Alexandrescu]

On 4/1/19 2:13 PM, Olivier FAURE wrote:
> tl;dr Keep strings and reference-counting separate.

You can't do that. New strings are created frequently by concatenation, 
they're best thought of as units - like int. They need to handle their 
own allocation, there's no two ways about it.

Autodecoding hurts "half" of the users - those who don't need it for the 
task. De-encapsulating string as an unstructured slice hurts the other 
"half" - those who need things like Unicode semantics, simple string 
manipulation, and good use of memory.

The view that strings must be char[] and that that's the "simple" choice 
has been a disaster for D. (I believe I have started to convince Walter 
of that.) It's simple like Go is simple - forcing complexity realities 
on users. Incomparably larger than autodecoding. It has forced virtually 
all string-processing code in D to use the garbage collector. Immutable 
was a short of adrenaline that prolonged the lifetime of strings by ten 
years. Before that, having unstructured strings that were also weirdly 
mutable was a double disaster.

We must have a UTF8 reference counted string that assists built-in 
literals of type immutable(char)[] and rally the entire language 
ecosystem around those.